Enhancing Reproducibility in Computational Workflows : Linking Semantics, Metadata, and Ontologies

Computational Science and Engineering (CSE) increasingly depends on complex, data-intensive workflows that encompass simulations and data processing pipelines. As metadata computations and parameter-driven simulations expand, the need for reliable, automated frameworks to reproduce computational tasks across different platforms has become essential. Computational workflows often define the multi-step processes used for data collection, preparation, analytics, predictivemodeling, and simulation. These workflows frequently involve interdependent tasks, diverse data sources, and heterogeneous computing environments. To manage such complexity and thereby improve reproducibility, workflow tools are nowadays integrated with semantic technologies. As part of the MaRDI consortium [1] on research data management (RDM) in mathematical sciences, we introduce MaRDIFlow [2], a lightweight, metadata-driven workflow tool for realizing and documenting computer-based experiments. In this framework, the workflow components are characterized by their input-output relationships, allowing associated metadata to be used interchangeably. We define suitable metadata and a low-level language for describing general CSE workflows at varying abstraction levels. Importantly, redundancy in representing models, code, and data is viewed as a strength, ensuring compatibility during execution and supporting reproducibility, the pillars of replicable and reusable research data. To enhance the semantic interoperability of our RDM tool, ontology integration is implemented via RESTful APIs or SPARQL endpoints, allowing real-time querying and validation of metadata during workflow construction and execution. This approach ensures semantic consistency across workflow components, so that both the components and their associated metadata are aligned with domain-specific standards. By leveraging such interfaces, workflow tools can dynamically retrievevocabulary terms, definitions, and relationships from ontologies, embedding them directly into the workflow metadata and supporting reproducibility. In this context, we integrate Voc4Cat [3], a domain-specific ontology and SKOS (Simple Knowledge Organization System) vocabulary developed by the NFDI4Cat consortium. When linked with MaRDIFlow, Voc4Cat acts as a backbone for annotating and defining metadata dependencies and relationships, supporting comprehensive workflow descriptions. Furthermore, we have implemented example use cases within the MaRDIFlow framework to demonstrate its applicability in executing CSE workflow descriptions. These examples showcase how MaRDIFlow, when integrated with domain-specific ontologies like Voc4Cat, leverages knowledge graphs to represent and query relationships across multi-layered workflow components, thereby enhancing semantic consistency and reproducibility.Going forward, we aim to extend MaRDIFlow to interface with additional ontologies and libraries developed within NFDI (Nationale Forschungsdateninfrastruktur), promoting broader interoperability across disciplines and contribute to NFDI’s mission to develop community driven research data infrastructure. References[1] MaRDI, Mathematical Research Data Initiative, 2021. [Online]. Available: https://www mardi4nfdi.de.[2] P. L. Veluvali, J. Heiland, and P. Benner, “MaRDIFlow: A CSE workflow framework for abstracting meta-data from FAIR computational experiments,” arXiv preprintarXiv:2405.00028, 2024.[3] D. Linke, N. Moustakas, M. Doerr, J. Schumann, and M. Götte, Nfdi4cat/voc4cat: Release 2024-12-17, version v2024-12-17, Dec. 2024. DOI : 10.5281/zenodo.14509593. [Online].Available: https://doi.org/10.5281/zenodo.14509593.